Speech-to-text Translation on MuST-C En-X (tst-COM)
30.1BLEU (German)SpeechUT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SpeechUTextra data=1.4k hours of speech and parallel MT data from WMT, task type=various mask-predict pre-training tasks2023.05 | 30.1 | 41.4 | 33.6 | 35 | |
| U2TT w/ DUButilize transcriptions=false, Scale=LARGE, DUB=true, Decoding strategy=Random sampling2023.05 | 26.2 | 35.3 | 30.4 | 30.6 | |
| U2TT w/ DUButilize transcriptions=false, Scale=BASE, DUB=true, Decoding strategy=Random sampling2023.05 | 25.8 | 34.7 | 30.2 | 30.2 | |
| Self-trainingutilize transcriptions=true2023.05 | 25.2 | 34.5 | — | — | |
| SpeechT5utilize transcriptions=true2023.05 | 25.2 | 35.3 | — | — | |
| MTLutilize transcriptions=true2023.05 | 23.9 | 33.1 | 28.6 | 28.5 | |
| Speechformerutilize transcriptions=true2023.05 | 23.6 | — | 28.5 | — | |
| Cascadedutilize transcriptions=true2023.05 | 23.6 | 33.8 | 28.7 | 28.7 | |
| E2E-ST-JTutilize transcriptions=true2023.05 | 23.1 | 32.8 | 27.5 | 27.8 | |
| Espnet STutilize transcriptions=true2023.05 | 22.9 | 32.8 | 28 | 27.9 | |
| NeurSTutilize transcriptions=true2023.05 | 22.8 | 33.3 | 27.4 | 27.8 | |
| Fairseq STutilize transcriptions=true2023.05 | 22.7 | 32.9 | 27.2 | 27.6 | |
| Revisit STutilize transcriptions=false2023.05 | 22 | 33.5 | 28 | 28.2 | |
| U2TTutilize transcriptions=false, Scale=LARGE, DUB=false2023.05 | 20.5 | 30.1 | 24.7 | 25.1 | |
| U2TTutilize transcriptions=false, Scale=BASE, DUB=false2023.05 | 20.4 | 30.3 | 25.3 | 25.3 | |
| Transformer-STutilize transcriptions=false2023.05 | 18 | 28.5 | 24.1 | 23.5 |